In all the database contains 5501 records, 4120 with pictures, 1072 with text only and 409 section or chapter headings. The text includes about 350,000 words.
Before the HTML files were created, each text record was pre-processed to insert cross-references and links to bibliographic citations. Paragraph identifiers, that would be converted to HTML targets, were also added. The scripts to generate the HTML files operate on each section as a unit, transferring the title and descriptive text for each picture record to one file and generating the links to the picture images. Similarly, the unillustrated text records in the section make up a text file. Separate scripts are used to generate the small files that display the high resolution versions of each image and the manuscript pages.

2. The Index
Compiling the index to Randle Holme's text, even in its edited form, proved to be one of the most onerous and technically difficult aspects of the project. As a first stage, the complete contents and description records were processed to produce a database of keywords, each accompanied by a context phrase and link. After correcting misprints, this table contained 354,000 records, referring to 24,000 keywords (of two letters or more) (Footnote 1). Because of Holme's arbitrary use of capitalisation, all the keywords were converted to lower case (including proper nouns). This lengthy list was reduced to 14,000 standardised keywords in which varieties of word form as well as spelling are combined. In general, verbs are given as infinitives (except where used adjectivally) and nouns are singular unless the plural form has a distinct meaning.
Finally, a selected keyword list was made, principally of those nouns and adjectives that seem most useful as index terms. Although the list was reduced only to 11,000 words, the total number of references dropped to 104,000; the number of keywords is expanded by the long lists of terms given by Holme, notably as beggar's cant and mariner's words; his Latin names of medicines have been eliminated from the selected word index to reduce its length.
The choice, both of standard spellings and of selected keywords, proved very difficult and we are not certain whether either process greatly enhances the usefulness of the index for readers. We therefore decided to create indexes from all three lists. Thus, additional usages of words exactly as cited by Randle Holme can be located in the 'original spelling' index, while even the commonest words can be found in the 'standard' index. This may be particularly useful for searching homonyms; for example, the noun can is buried among the verb forms. It was decided that individual words should either be completely included or excluded from the selected word index, so that only one index needs to be searched.
Using the Index
One technical aspect of the index should be noted. Because some of the link files are very large, they take a significant time to load. Clicking on a word opens the main text in a new window, so that the index files do not have to be reloaded before looking at another reference. Thus, to switch between index and text, either view both windows together, each reduced to about half the height of the full screen, or click alternately on the two Netscape windows on the taskbar. To find a particular word, click in the keyword frame and then type CNTL-F to search for the word.

3. Editing the drawings
The 247 pages of Holme's manuscripts that include drawings were scanned with the British Library's digital camera, to give a series of very high resolution TIFF files, each more than 20 MB in size. Generating the illustrations from these was lengthy but straightforward, using Adobe Photoshop. Each individual drawing was clipped and edited, if necessary, to remove confusing lines belonging to adjacent images; for a few very faint images, the contrast had to be increased. Two copies were then saved; the first was at full resolution, to display the maximum information about the image (footnote 2); the second, image was reduced in size to minimize its loading time. This version was resampled at 2/3rds resolution and down-sized to a standard height of 120 or 180 pixels, depending on the height of the original image; if the original after resampling was less than 120 pixels high, it was not further reduced.
For the illustrations of whole pages and the full page versions of the engraved plates, the originals were simply reduced to about 2/3rds resolution and saved as .jpg files; they can be clicked to view them at large scale. Smaller versions, scaled to a height of 600 pixels, were prepared for the manuscript page displays. Similarly, for the thumbnails used in the picture index, the plates were reduced to about one third resolution before clipping.
[Note: These reduced-size images are retained on the TATHS website, since it is not practicable to re-edit each drawing and replace them in the image folders.]
As the drawings were prepared, information about each was recorded in a database, particularly to identify its page source and location on the page; the locations are in a letter sequence, normally in columns unless RH clearly aligned the images in rows. Handling the pages of jottings proved difficult, as their subjects tend to be very heterogeneous and the images on these pages have been less fully indexed.

Identifying the images
The initial impression on glancing at the manuscript, of innumerable fascinating but confusing images, was totally confirmed as they were studied in detail. Only on a very few pages do the pictures follow the order on the final plates, while rather more pages have the images grouped by subject, though sometimes they relate to more than one plate. Within the Harleian volumes, linked topics are scattered, even to the extent of two halves of one original sheet being bound up in different volumes.
Despite this confusion, identifying the images was generally relatively easy, usually by matching them with the corresponding printed plate. For the sections whose plates have not survived (chapters 18b and 19b-21), groups of draft drawings could mostly be identified from the labels on the drawings. Only where these labels were absent did each drawing need to be compared very closely with the written texts. It was particularly gratifying towards the end of this phase of the project, when the number of sections lacking drawings became relatively small, to recognise on H.2034, f.210 the unlabelled drawings for the gardener's and the wax-chandler's tools, and to realise that the latter continued on H.2027, f.408v with the remarkable small group of torch-maker's tools, though we are defeated by the identity of the spiky tongs in the centre of this page, as perhaps was Randle Holme himself (if this is indeed his un-named tool number 77).
Several pages contain additional drawings clearly relating to their main theme, but not included on the engraved plates. These have been appended at appropriate places in the corresponding chapter. Ultimately, only Jeweller's and Brazier's tools and Masonry in chapter 20, and the Cook and Fisherman in chapter 21 remain without surviving illustrations.
In all, slightly more than 2,500 unique drawings have been identified, with 200 duplicates which provide additional information; another 800 duplicates are listed with the corresponding large-scale drawing, with links to their manuscript page; other duplicates exist, particularly on the jottings pages and on the plate-layout drawings which are not fully indexed. When alternative images exist, the final drawing has generally been chosen, sometimes supplemented by a draft version; images from the plate layout drawings have only been used if no others exist. In Book III, 690 drawings could not be located and the corresponding images from the engraved plates have been used. Some of the proof plates were re-worked before publication, with extra images inserted into blank spaces, so 100 illustrations have had to be taken from the published rather than the proof plates. For Book II, the original drawings of plants and animals have not been sought out; we felt that RH's re-drawing of a leopard or a cypress, certainly from a previous publication, was of less significance than his drawing of (say) a hat. Thus, in Book II, 980 illustrations are taken from the plates and only 12 from drawings.

Footnotes


1. At each stage of index preparation, misprints were found; although all those recognised have been corrected in the main text and in the keywords, a few remain in the context phrases. For the same reason, a very few of the links point to the text file rather than the picture file (or vice versa), because of last-minute corrections to the data, for which it was not possible to amend the index files.

2. A few very wide images had to be reduced slightly in size so that they can be viewed satisfactorily.